Reporting checklist

Use this as a copy/paste checklist for assignments, papers, and thesis chapters.

Always report (any SEM/CFA)

  • Software + version (R, lavaan, semTools, etc.)
  • Sample size (N) and any exclusions
  • Variables (which indicators for each factor; scales; whether ordinal)
  • Estimator (ML / MLR / WLSMV etc.) + justification
  • Missing data handling (FIML / MI / listwise) + justification
  • Model syntax (append or supplement)
  • Model identification choices (marker vs std.lv=TRUE, constraints, etc.)
  • Global fit: χ²(df), CFI/TLI, RMSEA (+ CI), SRMR
  • Key parameters: estimates + SE/CI + standardized solution (when meaningful)
  • Interpretation tied to the research question
  • Diagnostics: what you checked (residuals, MI/EPC) and what you changed (if anything)
  • Reproducibility: session info (and seed if simulation)

CFA-specific

  • Factor loadings (unstd + std), indicator residual variances
  • Factor variances/covariances
  • Reliability: ω (omega) and/or other chosen indices (with method stated)
  • Handling of cross-loadings / correlated residuals:
    • were any added?
    • what was the substantive rationale?
    • was this preregistered / cross-validated / replicated?

SEM (structural) specific

  • Structural paths (standardized + CI)
  • R² for endogenous variables
  • Indirect effects:
    • method (bootstrap recommended when appropriate)
    • effect + CI + interpretation (avoid “p-value only” reporting)
  • Clarify causal language:
    • observational vs experimental
    • identification assumptions (temporal ordering, confounding, etc.)

Multi-group / invariance

  • Groups (definition, Ns per group)
  • Invariance steps tested:
    • configural, metric, scalar (and strict if used)
  • What constraints were imposed at each step
  • Decision rules used (e.g., ΔCFI / ΔRMSEA thresholds) and references for your rule
  • If partial invariance:
    • which parameters freed
    • rationale (substantive + statistical)
    • what comparisons remain interpretable (e.g., latent means only after (partial) scalar)

Ordinal indicators

  • Which variables treated as ordered
  • Estimator and link (e.g., WLSMV / probit-type default framework)
  • Thresholds vs intercepts (make clear what was constrained in invariance)
  • Fit interpretation caveat: note if thresholds or residual structures drive misfit
  • Sensitivity check (optional but strong): treat as continuous vs ordered and compare conclusions

Longitudinal SEM

  • Time points and spacing
  • Whether you tested longitudinal measurement invariance
  • How you handled correlated uniqueness (same items across time)
  • For growth models:
    • coding of slope loadings (0,1,2… or actual time)
    • random effects included (intercept/slope variance)
  • For cross-lagged models:
    • clarify whether you used CLPM vs alternatives (if relevant)
    • interpret cross-lagged paths cautiously

Clustered / multilevel

  • Clustering variable (e.g., class, clinic, therapist) and cluster sizes
  • Strategy used:
    • cluster-robust SE (and why sufficient), or
    • explicit multilevel model (within/between)
  • If multilevel:
    • which parameters vary within vs between
    • ICC motivation (optional but helpful)

A short “results paragraph” template

“We fit a [CFA/SEM] using lavaan (version …) with estimator … . Missing data were handled using … . Model fit was [acceptable/problematic] (χ²(df)=…, CFI=…, RMSEA=… [CI], SRMR=…). The key effect of interest was … (β=…, 95% CI […, …]). Diagnostics indicated … ; we [did/did not] respecify the model. Results suggest …, with limitations including … .”

Important

Rule of thumb
If you changed the model after looking at the data, state exactly what you changed and why. “Transparency beats perfection.”